# /srv/irclogs.ubuntu.com/2008/02/06/#ubuntu-installer.txt

tagman tagman CIA-24 How do I get past the installer being stuck at 15%? 01:51 Anybody there? 01:55 ubiquity: evand * r2438 ubiquity/ (6 files in 5 dirs): 03:07 ubiquity: * Validate the boot device string that the user provides on the 03:07 ubiquity:  advanced page. 03:07 ubiquity: * Change the boot device Entry to a ComboBox that displays suitable targets. 03:07 ugh, pyqt is quite evil. 06:30 trying to implement the equivalent of that above commit for qt? 06:33 yeah, and weird language gotchas like having to str() qstrings 06:36 I just want a model with two columns, is that so much to ask for? 06:36 apparently :) 06:37 haha 06:37 you see why i opted for the mythbuntu frontend to derive off the gtk one then?  I played with the qt designer for a little bit and decided it would be way to rough to work from 06:38 I used to know Qt quite well :/ 06:38 but yes, my brain can handle GTK much better 06:39 base-installer: soren * r322 base-installer/ (debian/changelog kernel/i386.sh): Add support for AMD CPU family 16 on i386. 07:56 speaking of GTK, if anyone knows how to implement a multi-column list in GTK where the columns flow newspaper-style rather than meaning different things, that'd be great 09:08 that is, it's only multi-column in that when it runs out of vertical space in one column it goes on to the next 09:08 cjwatson: I don't think there's a widget for that in GTK, but I guess you could use a regular TreeView and call gtk_tree_view_get_visible_* to determine how many rows are showing and then do it manually. 09:29 === cjwatson_ is now known as cjwatson soren: mm, that's the best I've come up with - it'll be nasty on resize though 10:40 I think I'll ask gtk-list if anyone has any sample code 10:44 how to be a member in ubuntu-installer@lists.ubuntu.com 11:38 ? 11:38 sridhar: https://lists.ubuntu.com/mailman/listinfo/ubuntu-installer 11:53 cjwatson: thanks 12:06 superm1: I'm looking at your preservehome branch at the moment, and comparing it with the adduser source 12:09 superm1: it seems to me that if you pass --gid to adduser it expects the group to exist already (although this isn't particularly documented) 12:09 superm1: did you test this and confirm it to work? 12:09 superm1: I think that you probably need to call addgroup explicitly if GIDOPT is set 12:10 does d-i run any scripts _before_ inittab is read? 12:23 outside the initrd 12:23 d-i runs entirely from its initrd so that question has no meaning ... 12:23 heh 12:24 rootskel/src/init 12:24 that does some stuff and then execs busybox init which reads inittab 12:24 :q 12:24 (oops) 12:24 basically just mounts /proc and creates some devices 12:25 oh, and mounts /dev to put said devices in, of course 12:25 I'm (again) thinking how it would be possible to disable the "convenience shells" run from inittab. maybe it would be too fugly to read a kernel parameter in rootskel/src/init and echo those tty-lines in a normal situation? 12:27 you could read a kernel parameter and have it sed -i /etc/inittab 12:27 to eliminate them 12:27 remind me why you're disabling them? 12:27 unattended installations on a public place 12:28 ok, hope you've locked the bootloader 12:28 sure 12:28 that's easy :) 12:28 netboot supports passwords now 12:28 I think honouring a 'noshell' parameter or similar would be fine 12:28 but the place to do it is in src/init? 12:29 yeah 12:29 ok cool, I'll make a debdiff 12:29 feel free to just commit and upload 12:30 heh, ok :) 12:30 bzr branch in the usual place 12:30 I'm so excited that Matt made all those fancy patches for syslog and sshkey-authentication for network-console :) 12:31 d-i will simply rock 12:31 oh, it already does btw 12:31 but even more 12:31 === cjwatson_ is now known as cjwatson 12:31 I'm so excited that Matt made all those fancy patches for syslog and sshkey-authentication for network-console :) 12:43 was the last thing I saw 12:43 tjaalton: yeah, it's pretty cool 12:43 14:31 < tjaalton> d-i will simply rock 12:43 14:31 < tjaalton> oh, it already does btw 12:43 14:31 < tjaalton> but even more 12:43 :) 12:43 haha 12:45 cjwatson, I'd like to add hooks similar to lupin/casper into d-i, what would be a good place? initrd, rootskel or udeb? 13:10 the reason is to support distros that do not ship a live CD and to make it possible to have netinstall capabilities in the future 13:10 that depends. hooks for what? 13:11 1 hook to set a local ISO skiping the stage where an ISO is looked for (mount it as /cdrom?) 13:11 1 hook to do initrd preseeding using a file on local device 13:12 possibly keeping the same boot arguments 13:12 as in casper 13:12 haven't we done this before? this should use hd-media 13:17 I do not want to add this to the CD initrd 13:17 before I was using the initrd since I needed to do overrides of the ISO 13:18 of the d-i installer^ 13:18 d-i has multiple initrd variants 13:18 but I do not think it is necessary 13:18 anymore 13:18 I'd go for either rootskel/src/lib/debian-installer.d or a udeb to be called before initrd preseeding 13:19 rootskel is completely and utterly inappropriate 13:19 can you remind me when initrd preseeding kicks in (web interface is a pain to look code for) 13:20 that seems to leave udeb route 13:20 using a local ISO is already implemented by iso-scan 13:20 initrd preseeding using a file on a local device is a contradiction in terms 13:20 the purpose of initrd preseeding is to be entirely local to the initrd 13:20 I think I used to do to seed locale/layoutcode... but can do that via boot parameters anyway 13:21 indeed so 13:21 file-preseed runs after iso-scan and can already be used in hd-media mode to preseed from a file on a local device 13:22 it would be good then to specify the path/name of the ISO 13:22 would be good then to specify the path/name of the ISO/preseed 13:22 which takes precedence over standard paths 13:22 it would be reasonable for iso-scan to support preseeding of iso-scan/filename (which it already uses as an output) 13:23 it is already possible to set paramaters to look for a specific local path for preseeding correct? 13:24 yes 13:24 so that /path/to/preseed.cfg is looked for in all devices 13:24 err, no 13:25 only on things that are already mounted 13:25 and that depends on... 13:25 I am very cautious of things that run around mounting everything in d-i 13:25 they have been known to cause data loss 13:26 (e.g. hibernated partitions) 13:26 only a temporary mount is required, enough to copy over preseed (and possibly other files) then umounting 13:26 it's difficult to specify the linux device from windows 13:27 I know, but that doesn't really help matters 13:27 mounting a hibernated partition that uses a journalled filesystem is enough to corrupt it 13:27 also a ro mount? 13:28 yes 13:28 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=417407 suggests a scheme for avoiding this 13:28 Debian bug 417407 in os-prober "os-prober: protect partitions with "blockdev --setro"" [Important,Open] 13:28 doesn't the iso-scan mount partitions anyway 13:28 but one does have to be pretty careful 13:28 IIRC 13:28 that would be not much different 13:28 hmm, yeah, I suppose iso-scan does have that problem; yet another reason it is not in the cdrom initrd :) 13:28 the boot arguments currently used by casper would not be accepted into upstream d-i 13:29 I did warn you about this ages ago ;) 13:30 on my side the situation is that I have C:\path\to\ISO and C:\path\to\preseed.cfg and need to convey this info over to the installer 13:30 all d-i boot arguments used by udebs must correspond to debconf questions and must be properly namespaced 13:30 so if you want this in d-i there's going to have to be a transition 13:30 if you have any good idea on that I'd be more than happy to implement it 13:30 things implemented by iso-scan must generally be iso-scan/foo, etc. 13:31 casper of course has to use the initrd, while that is not required for d-i (as mentioned above), not sure if that helps matters 13:31 I can do iso-scan mods, that means though that I have to use hdmedia initrd as opposed to the one found in the alternate ISO, correct? 13:33 or do a netinstall skiiping isoscan (would still need preseed-scan though) 13:33 on a different topic, I do not know how I missed it for so long, but there is an nsis package in the archives 13:35 this greatly simplifies the build process 13:35 only thing is that I'd need v2.34 (now 2.33) since new code uses nsdialog interface which is much saner than previous implementations 13:36 at present, iso-scan fundamentally conflicts with the normal cdrom boot process 13:39 it might be possible to fix that in some way, but I haven't thought about it much and cannot guarantee that it's straightforward 13:39 can we have either one or the other accoring to boot arg maybe? 13:40 no 13:40 well, not right now 13:40 that would obviously be roughly the way to fix it but it needs some thought from a d-i developer to get it right 13:41 hdmedia initrd + iso would still work though (assuming I can push the changes to hdmedia) 13:42 iirc hdmedia also works for netinstallations 13:42 I'd be happy with that too 13:43 network installations should generally use the netboot initrd 13:44 hd-media doesn't include net-retriever 13:45 haven't looked at it in a while but thought that much functionality of netboot is also in hdmedia 13:45 net-retriever is kind of important :) 13:45 you can't get udebs from the network without it 13:45 then I didn't remember correctly... 13:46 any reason to have different initrds as opposed of having boot switches? wouldn't think that the size would change dramatically 13:47 they conflict 13:49 and actually the size is quite important and yes it does make a difference 13:49 different boot methods require different sets of kernel modules 13:49 the netboot initrd doesn't need block device drivers, for instance; it can retrieve those later 13:50 similarly the cdrom and hd-media initrds can save quite a bit of space by leaving out network drivers 13:50 I am fine with it, it only makes it slightly inconvenient for people that want to use wubi in offline mode 13:55 since on top of the ISO/CD they also need to get an appropriate initrd (but only in some cases) 13:55 I'm not saying it's perfect, just that there isn't a quick fix and that it needs some design work, that's all 13:56 to conclude then I'd gess the best route might be to add an iso-scan/path parameter that will affect hdmedia 13:56 iso-scan/filename (since the debconf question already exists) 13:57 yes, that seems like a fairly obvious first step 13:57 as for the preseed fetching I am a bit more cautious, I'd also need something similar to the above 13:57 does current local-device-preseed-fetching only affects hdmedia? 13:58 in case I can simply patch that 13:58 well, there isn't really such a thing right now 13:58 file-preseed knows how to fetch a preseed file from a specified path 13:59 (about to be on the phone for two hours, sigh) 13:59 can we change (in a backward compatible way) the path syntax so that I can specify ANYDEVICE/path/to/preseed.cfg? 14:00 I think it would be better to add a second question which says to do a device search, rather than to try to crowbar it into the syntax 14:01 see also ^ re nsis 2.34, I'll need to nag someone to have that in the archives... 14:08 cjwatson the above is certainly ok for hdmedia, but not for netinstall (if block device drivers are not there I am not sure how I can preseed it) 14:27 That is only an issue for dedicated partition installation, you mentioned that if I use a local ISO with hdmedia, repartitioning is an issue, does that still hold true? 14:28 yep 14:57 I assume that using "iso-scan/filename" also in casper is not acceptable since there is no iso-scan package. But would be nice to standardize on that 15:04 cjwatson, I didn't actually get to test it to confirm it works.  I'll see if I can give that a go. 15:05 same with debian-installer/custom-installation (now in casper) 15:05 wich includes preseed 15:06 rootskel uploaded, hope I got it right the first time 15:14 iso-scan/filename in casper would be OK; casper already abuses namespaces somewhat for compatibility 15:23 ok I'll change that, I like it better than find_iso anyway 15:26 what about debian-installer/custom-installation within d-i? maybe scan-custom-installation/foldername? 15:27 I am on the phone but will get back to you later 15:27 sure 15:27 preliminary iso-scan "patch" to support preseeded filename http://paste.ubuntu-nl.org/54997/ 16:41 please follow local style in terms of indentation 16:51 (indent is one tab except for case patterns which are half-indent, i.e. four spaces, to keep the depth reasonable) 16:52 I think it would be better to just whack $DEVS and go through the normal code path, to avoid duplicating that code 16:53 Bug #189630 summarize above discussion 16:53 Launchpad bug 189630 in lupin "d-i/hdmedia should have hooks similar to the ones in lupin/casper" [Undecided,New] https://launchpad.net/bugs/189630 16:53 and I really think this ought to be controlled by a separate preseeded variable which is off by default 16:53 you mean iso-scan/filename? 16:54 I don't mean$DEVS, sorry, I mean the output of the directory-scanning code in the inner loop 16:54 no 16:54 well, er 16:54 I just factored out try_dir since the same block would occur 3 times 16:55 oh, um, ok, maybe I misunderstood 16:55 could you send an actual patch once you can 16:55 can do that tonight 16:55 even a debdiff at that :P 16:55 I think rather than try_preseeded_filename you should just insert this into the loop that goes 'for iso in $dir/*.iso$dir/*.ISO; do' 16:56 and have it check iso-scan/filename first and only if it's not set then do that scan 16:56 then it doesn't really need any refactoring AFAICS 16:57 then it should go in "for dev in devs", since I only need to check once per dev 16:57 actually, probably should be around 'for dir in . *; do' 16:57 right 16:57 sorry, hard to read code while chairing a meeting ;) 16:58 np 16:58 xivulon: 17:03 11:00:25 < TheMuso> evand: I think atm, a series of radio buttons, as thats all that can be done from gfxboot, and I don't  think there is a need for more than one profile needed at the same time. 17:03 11:00:43 < evand> TheMuso: fantastic, thanks 17:03 11:00:48 < cjwatson> I actually implemented checkbox support in gfxboot-theme-ubuntu recently 17:03 11:00:53 < cjwatson> in a hideously crude way but never mind 17:03 11:01:01 < cjwatson> ("x  " FTW) 17:03 11:01:11 < evand> haha 17:04 11:01:12 < TheMuso> cjwatson: I still don't think it makes sense in this context. 17:04 I'm having an odd, odd problem trying to build d-i.. 17:09 The symptom: 17:09 During get-packages, I get: 17:09 cyberorc: Helt sikkert. Har jævnt travlt, men det er ca. som forventet.The following packages have unmet dependencies: console-tools-udeb: Depends: libconsole (= 1:0.2.3dbs-65ubuntu6) but it is not installable 17:09 E: Broken packages 17:09 Whoops. 17:09 Only this bit: The following packages have unmet dependencies: console-tools-udeb: Depends: libconsole (= 1:0.2.3dbs-65ubuntu6) but it is not installable 17:09 * soren curses his mouse wheel 17:09 evand thanks, I am not convinced that 1 profile is enough but if the TheMuso says so, it's fine with me 17:11 I can't imagine what would make libconsole uninstallable. 17:13 grub-installer: evand * r724 grub-installer.ubuntu/ (debian/changelog grub-installer): 17:21 grub-installer: * Generate a device.map when bootdev_directory is set (loop 17:21 grub-installer:  installations) (LP: #188460). Thanks Agostino Russo. 17:21 cjwatson: Hm... the sources.list.udeb doesn't list a regular main, only the d-i one.. How is it supposed to get its hands on libconsole? Is it supposed to be a build-dependency of d-i? 17:22 grub-installer: evand * r725 ubuntu/debian/changelog: releasing version 1.27ubuntu3 17:23 cjwatson: Or is the main archive's absence from sources.list.udeb a bug? 17:23 err, you might just need to have it installed on the build system 17:34 the main archive should definitely not be in sources.list.udeb 17:34 actually, why is console-tools-udeb depending on libconsole rather than libconsole-udeb? that's a bug 17:35 I thought I fixed that ages ago 17:35 I bet dpkg broke that on me 17:35 udeb: libconsole 0 libconsole-udeb (= 1:0.2.3dbs-65ubuntu6) 17:36 shlibs *look* right ... 17:36 It built correctly on the buildd's as well. How odd. 17:44 Ah, new console-tools from the day before yesterday. 17:45 urr. any chance you could figure out why console-tools-udeb is getting the wrong dependency, despite the shlibs? 17:47 the change in console-tools isn't relevant AFAIK; it's a build system thing 17:49 WEll, I guess I'm going to have to. 17:49 cjwatson: Sure, I was just wondering why it worked a week ago. 17:49 cjwatson: and that explains it. 17:49 it hadn't been built since gutsy, so I expect it needs tweaked to cope with dpkg-shlibdeps changes 17:50 it was doing some kind of nasty stuff 17:50 Will you be around for the next couple of hours or later this evening? 17:51 My crystal ball tells me I'll find the issue, but won't be able to tell if it's dpkg or console-tools that's doing the wrong thing. 17:51 s/dpkg/dpkg and friends/ 17:51 cjwatson: Hrm... There's a shlibs.local, that says "libconsole0libconsole (= ${Source-Version}) 17:59 Erk.. That looks mangled here. 17:59 cjwatson: Hrm... There's a shlibs.local, that says "libconsole 0 libconsole (=${Source-Version})". 17:59 cjwatson: dpkg-shlibdeps doesn't mention anything about udeb. I have literally no historical context, so I'm not sure where to point my finger. 18:00 cjwatson: Adding the udeb: lines to shlibs.local fixes it.. 18:06 * soren goes to dinner. bbl. 18:07 soren: later as in much later 18:08 not the next couple of hours 18:08 urr 18:09 I *think* that shlibs.local should just go away 18:09 would be worth seeing what happens if you just rm the thing 18:09 cjwatson_: Yeah, everything looks lovely without shlibs.local AFAICS. 20:54 cjwatson_: Thanks for your input. 20:54 great, thanks 20:59 === cjwatson_ is now known as cjwatson I like people figuring out d-i build failures for me ;) 20:59 :) 21:02 I'm curious why it suddenly broke, though. 21:02 That shlibs.local has been the for a looong time, apparantly. 21:02 cjwatson: d-i's build/config/common still says "SUITE =? gutsy". I doubt that's intentional but somehow I feel I should probably ask to be sure :) 21:43 yeah, doesn't actually matter because debian/rules overrides it, but should be fixed; go ahead 21:45 suddenly broke> dpkg-shlibdeps was practically rewritten ... 21:45 Ah, yes. 21:48 cjwatson: Ok, a bit of assistance... I'm working on the iscsi thing again. I've made changes to disk-detect and added a new udeb... How do I most easily test these things? the mini.iso d-i can build for me doesn't have the udeb's so that's not as easy as I had hoped.. 22:16 My apologies if this is already explained somewhere.. 22:16 localudebs? 22:20 evand: Yeah, but they don't land on the mini.iso, do they? 22:20 I thought they did, but it's been a while. 22:20 or am I being a muppet? 22:20 It's quite conceivable that I am. 22:21 heh 22:21 Nope. 22:22 argh, not sure then 22:22 aha! 22:24 EXTRAUDEBS ftw, it seems. 22:24 for adding a new udeb, localudebs or EXTRAUDEBS is about right 22:28 you might need to prod build/pkg-lists/whatever 22:28 for simpler changes it's often sane to just edit things on the fly in the running installer 22:29 that doesn't work if you add debconf templates though 22:29 I do. 22:31 But the mini iso should do fine? 22:32 Or do I need to dive into debian-cd, too? 22:32 mini.iso should do fine provided it's the cdrom one 22:36 (mini.iso is a delivery format rather than something that defines what udebs go on the image) 22:36 if cdrom doesn't work, monolithic is the standard big hammer 22:37 you'll need to edit build/config/i386.cfg or similar to uncomment monolithic, and then run 'make' in build/ to get the right target name 22:37 probably rebuild_monolithic 22:37 Ah... adding them to EXTRAS and cleaning and rebuilding got them into the initrd. That was what I was looking for, I guess. 22:38 monolithic builds all available udebs into the image 22:38 Ah, ok. 22:38 except that it doesn't need a retriever 22:38 because it already has it all in the initrd 22:38 obviously fairly giant 22:38 I can imagine :) 22:39 cjwatson: I remember you said I could define /proc/cmdline aliases... Where was that again? 22:41 preseed_aliases 22:41 in the preseed source 22:41 I think it's /lib/preseed/preseed_aliases in the initrd 22:41 right, early(ish) night time, I think 22:42 Alright. I think I have plenty to work with now. Thanks for your help. 22:43 no problem 22:43 cjwatson: Have you gone yet? 23:04 cjwatson re iso-scan: http://paste.ubuntu-nl.org/55035/ 23:26 haven't tested it though 23:26 howdy 23:55 howdy 23:55 when I use debuild I get "secret key not available" 23:56 the key is in my keyring, do I need to register it homehow? 23:56

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!